util/main-loop.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu/osdep.h"
  26 #include "qapi/error.h"
  27 #include "qemu/cutils.h"
  28 #include "qemu/timer.h"
  29 #include "qemu/sockets.h"       // struct in_addr needed for libslirp.h
  30 #include "sysemu/qtest.h"
  31 #include "sysemu/cpus.h"
  32 #include "slirp/libslirp.h"
  33 #include "qemu/main-loop.h"
  34 #include "block/aio.h"
  35
  36 #ifndef _WIN32
  37
  38 /* If we have signalfd, we mask out the signals we want to handle and then
  39  * use signalfd to listen for them.  We rely on whatever the current signal
  40  * handler is to dispatch the signals when we receive them.
  41  */
  42 static void sigfd_handler(void *opaque)
  43 {
  44     int fd = (intptr_t)opaque;
  45     struct qemu_signalfd_siginfo info;
  46     struct sigaction action;
  47     ssize_t len;
  48
  49     while (1) {
  50         do {
  51             len = read(fd, &info, sizeof(info));
  52         } while (len == -1 && errno == EINTR);
  53
  54         if (len == -1 && errno == EAGAIN) {
  55             break;
  56         }
  57
  58         if (len != sizeof(info)) {
  59             printf("read from sigfd returned %zd: %m\n", len);
  60             return;
  61         }
  62
  63         sigaction(info.ssi_signo, NULL, &action);
  64         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  65             sigaction_invoke(&action, &info);
  66         } else if (action.sa_handler) {
  67             action.sa_handler(info.ssi_signo);
  68         }
  69     }
  70 }
  71
  72 static int qemu_signal_init(void)
  73 {
  74     int sigfd;
  75     sigset_t set;
  76
  77     /*
  78      * SIG_IPI must be blocked in the main thread and must not be caught
  79      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  80      * not catch it reliably.
  81      */
  82     sigemptyset(&set);
  83     sigaddset(&set, SIG_IPI);
  84     sigaddset(&set, SIGIO);
  85     sigaddset(&set, SIGALRM);
  86     sigaddset(&set, SIGBUS);
  87     /* SIGINT cannot be handled via signalfd, so that ^C can be used
  88      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
  89      * SIGTERM are also handled asynchronously, even though it is not
  90      * strictly necessary, because they use the same handler as SIGINT.
  91      */
  92     pthread_sigmask(SIG_BLOCK, &set, NULL);
  93
  94     sigdelset(&set, SIG_IPI);
  95     sigfd = qemu_signalfd(&set);
  96     if (sigfd == -1) {
  97         fprintf(stderr, "failed to create signalfd\n");
  98         return -errno;
  99     }
 100
 101     fcntl_setfl(sigfd, O_NONBLOCK);
 102
 103     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 104
 105     return 0;
 106 }
 107
 108 #else /* _WIN32 */
 109
 110 static int qemu_signal_init(void)
 111 {
 112     return 0;
 113 }
 114 #endif
 115
 116 static AioContext *qemu_aio_context;
 117 static QEMUBH *qemu_notify_bh;
 118
 119 static void notify_event_cb(void *opaque)
 120 {
 121     /* No need to do anything; this bottom half is only used to
 122      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 123      */
 124 }
 125
 126 AioContext *qemu_get_aio_context(void)
 127 {
 128     return qemu_aio_context;
 129 }
 130
 131 void qemu_notify_event(void)
 132 {
 133     if (!qemu_aio_context) {
 134         return;
 135     }
 136     qemu_bh_schedule(qemu_notify_bh);
 137 }
 138
 139 static GArray *gpollfds;
 140
 141 int qemu_init_main_loop(Error **errp)
 142 {
 143     int ret;
 144     GSource *src;
 145     Error *local_error = NULL;
 146
 147     init_clocks(qemu_timer_notify_cb);
 148
 149     ret = qemu_signal_init();
 150     if (ret) {
 151         return ret;
 152     }
 153
 154     qemu_aio_context = aio_context_new(&local_error);
 155     if (!qemu_aio_context) {
 156         error_propagate(errp, local_error);
 157         return -EMFILE;
 158     }
 159     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 160     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 161     src = aio_get_g_source(qemu_aio_context);
 162     g_source_set_name(src, "aio-context");
 163     g_source_attach(src, NULL);
 164     g_source_unref(src);
 165     src = iohandler_get_g_source();
 166     g_source_set_name(src, "io-handler");
 167     g_source_attach(src, NULL);
 168     g_source_unref(src);
 169     return 0;
 170 }
 171
 172 static int max_priority;
 173
 174 #ifndef _WIN32
 175 static int glib_pollfds_idx;
 176 static int glib_n_poll_fds;
 177
 178 static void glib_pollfds_fill(int64_t *cur_timeout)
 179 {
 180     GMainContext *context = g_main_context_default();
 181     int timeout = 0;
 182     int64_t timeout_ns;
 183     int n;
 184
 185     g_main_context_prepare(context, &max_priority);
 186
 187     glib_pollfds_idx = gpollfds->len;
 188     n = glib_n_poll_fds;
 189     do {
 190         GPollFD *pfds;
 191         glib_n_poll_fds = n;
 192         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 193         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 194         n = g_main_context_query(context, max_priority, &timeout, pfds,
 195                                  glib_n_poll_fds);
 196     } while (n != glib_n_poll_fds);
 197
 198     if (timeout < 0) {
 199         timeout_ns = -1;
 200     } else {
 201         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 202     }
 203
 204     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 205 }
 206
 207 static void glib_pollfds_poll(void)
 208 {
 209     GMainContext *context = g_main_context_default();
 210     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 211
 212     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 213         g_main_context_dispatch(context);
 214     }
 215 }
 216
 217 #define MAX_MAIN_LOOP_SPIN (1000)
 218
 219 static int os_host_main_loop_wait(int64_t timeout)
 220 {
 221     GMainContext *context = g_main_context_default();
 222     int ret;
 223     static int spin_counter;
 224
 225     g_main_context_acquire(context);
 226
 227     glib_pollfds_fill(&timeout);
 228
 229     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 230      * the I/O thread, this can lead to starvation of the BQL such that the
 231      * VCPU threads never run.  To make sure we can detect the later case,
 232      * print a message to the screen.  If we run into this condition, create
 233      * a fake timeout in order to give the VCPU threads a chance to run.
 234      */
 235     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 236         static bool notified;
 237
 238         if (!notified && !qtest_enabled() && !qtest_driver()) {
 239             fprintf(stderr,
 240                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 241                     MAX_MAIN_LOOP_SPIN);
 242             notified = true;
 243         }
 244
 245         timeout = SCALE_MS;
 246     }
 247
 248     if (timeout) {
 249         spin_counter = 0;
 250         qemu_mutex_unlock_iothread();
 251     } else {
 252         spin_counter++;
 253     }
 254
 255     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 256
 257     if (timeout) {
 258         qemu_mutex_lock_iothread();
 259     }
 260
 261     glib_pollfds_poll();
 262
 263     g_main_context_release(context);
 264
 265     return ret;
 266 }
 267 #else
 268 /***********************************************************/
 269 /* Polling handling */
 270
 271 typedef struct PollingEntry {
 272     PollingFunc *func;
 273     void *opaque;
 274     struct PollingEntry *next;
 275 } PollingEntry;
 276
 277 static PollingEntry *first_polling_entry;
 278
 279 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 280 {
 281     PollingEntry **ppe, *pe;
 282     pe = g_malloc0(sizeof(PollingEntry));
 283     pe->func = func;
 284     pe->opaque = opaque;
 285     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 286     *ppe = pe;
 287     return 0;
 288 }
 289
 290 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 291 {
 292     PollingEntry **ppe, *pe;
 293     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 294         pe = *ppe;
 295         if (pe->func == func && pe->opaque == opaque) {
 296             *ppe = pe->next;
 297             g_free(pe);
 298             break;
 299         }
 300     }
 301 }
 302
 303 /***********************************************************/
 304 /* Wait objects support */
 305 typedef struct WaitObjects {
 306     int num;
 307     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 308     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 309     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 310     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 311 } WaitObjects;
 312
 313 static WaitObjects wait_objects = {0};
 314
 315 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 316 {
 317     WaitObjects *w = &wait_objects;
 318     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 319         return -1;
 320     }
 321     w->events[w->num] = handle;
 322     w->func[w->num] = func;
 323     w->opaque[w->num] = opaque;
 324     w->revents[w->num] = 0;
 325     w->num++;
 326     return 0;
 327 }
 328
 329 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 330 {
 331     int i, found;
 332     WaitObjects *w = &wait_objects;
 333
 334     found = 0;
 335     for (i = 0; i < w->num; i++) {
 336         if (w->events[i] == handle) {
 337             found = 1;
 338         }
 339         if (found) {
 340             w->events[i] = w->events[i + 1];
 341             w->func[i] = w->func[i + 1];
 342             w->opaque[i] = w->opaque[i + 1];
 343             w->revents[i] = w->revents[i + 1];
 344         }
 345     }
 346     if (found) {
 347         w->num--;
 348     }
 349 }
 350
 351 void qemu_fd_register(int fd)
 352 {
 353     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 354                    FD_READ | FD_ACCEPT | FD_CLOSE |
 355                    FD_CONNECT | FD_WRITE | FD_OOB);
 356 }
 357
 358 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 359                         fd_set *xfds)
 360 {
 361     int nfds = -1;
 362     int i;
 363
 364     for (i = 0; i < pollfds->len; i++) {
 365         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 366         int fd = pfd->fd;
 367         int events = pfd->events;
 368         if (events & G_IO_IN) {
 369             FD_SET(fd, rfds);
 370             nfds = MAX(nfds, fd);
 371         }
 372         if (events & G_IO_OUT) {
 373             FD_SET(fd, wfds);
 374             nfds = MAX(nfds, fd);
 375         }
 376         if (events & G_IO_PRI) {
 377             FD_SET(fd, xfds);
 378             nfds = MAX(nfds, fd);
 379         }
 380     }
 381     return nfds;
 382 }
 383
 384 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 385                          fd_set *wfds, fd_set *xfds)
 386 {
 387     int i;
 388
 389     for (i = 0; i < pollfds->len; i++) {
 390         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 391         int fd = pfd->fd;
 392         int revents = 0;
 393
 394         if (FD_ISSET(fd, rfds)) {
 395             revents |= G_IO_IN;
 396         }
 397         if (FD_ISSET(fd, wfds)) {
 398             revents |= G_IO_OUT;
 399         }
 400         if (FD_ISSET(fd, xfds)) {
 401             revents |= G_IO_PRI;
 402         }
 403         pfd->revents = revents & pfd->events;
 404     }
 405 }
 406
 407 static int os_host_main_loop_wait(int64_t timeout)
 408 {
 409     GMainContext *context = g_main_context_default();
 410     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 411     int select_ret = 0;
 412     int g_poll_ret, ret, i, n_poll_fds;
 413     PollingEntry *pe;
 414     WaitObjects *w = &wait_objects;
 415     gint poll_timeout;
 416     int64_t poll_timeout_ns;
 417     static struct timeval tv0;
 418     fd_set rfds, wfds, xfds;
 419     int nfds;
 420
 421     g_main_context_acquire(context);
 422
 423     /* XXX: need to suppress polling by better using win32 events */
 424     ret = 0;
 425     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 426         ret |= pe->func(pe->opaque);
 427     }
 428     if (ret != 0) {
 429         g_main_context_release(context);
 430         return ret;
 431     }
 432
 433     FD_ZERO(&rfds);
 434     FD_ZERO(&wfds);
 435     FD_ZERO(&xfds);
 436     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 437     if (nfds >= 0) {
 438         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 439         if (select_ret != 0) {
 440             timeout = 0;
 441         }
 442         if (select_ret > 0) {
 443             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 444         }
 445     }
 446
 447     g_main_context_prepare(context, &max_priority);
 448     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 449                                       poll_fds, ARRAY_SIZE(poll_fds));
 450     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 451
 452     for (i = 0; i < w->num; i++) {
 453         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 454         poll_fds[n_poll_fds + i].events = G_IO_IN;
 455     }
 456
 457     if (poll_timeout < 0) {
 458         poll_timeout_ns = -1;
 459     } else {
 460         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 461     }
 462
 463     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 464
 465     qemu_mutex_unlock_iothread();
 466     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 467
 468     qemu_mutex_lock_iothread();
 469     if (g_poll_ret > 0) {
 470         for (i = 0; i < w->num; i++) {
 471             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 472         }
 473         for (i = 0; i < w->num; i++) {
 474             if (w->revents[i] && w->func[i]) {
 475                 w->func[i](w->opaque[i]);
 476             }
 477         }
 478     }
 479
 480     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 481         g_main_context_dispatch(context);
 482     }
 483
 484     g_main_context_release(context);
 485
 486     return select_ret || g_poll_ret;
 487 }
 488 #endif
 489
 490 int main_loop_wait(int nonblocking)
 491 {
 492     int ret;
 493     uint32_t timeout = UINT32_MAX;
 494     int64_t timeout_ns;
 495
 496     if (nonblocking) {
 497         timeout = 0;
 498     }
 499
 500     /* poll any events */
 501     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 502     /* XXX: separate device handlers from system ones */
 503 #ifdef CONFIG_SLIRP
 504     slirp_pollfds_fill(gpollfds, &timeout);
 505 #endif
 506
 507     if (timeout == UINT32_MAX) {
 508         timeout_ns = -1;
 509     } else {
 510         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 511     }
 512
 513     timeout_ns = qemu_soonest_timeout(timeout_ns,
 514                                       timerlistgroup_deadline_ns(
 515                                           &main_loop_tlg));
 516
 517     ret = os_host_main_loop_wait(timeout_ns);
 518 #ifdef CONFIG_SLIRP
 519     slirp_pollfds_poll(gpollfds, (ret < 0));
 520 #endif
 521
 522     /* CPU thread can infinitely wait for event after
 523        missing the warp */
 524     qemu_start_warp_timer();
 525     qemu_clock_run_all_timers();
 526
 527     return ret;
 528 }
 529
 530 /* Functions to operate on the main QEMU AioContext.  */
 531
 532 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 533 {
 534     return aio_bh_new(qemu_aio_context, cb, opaque);
 535 }